The CALIBER Research Platform Using large-scale linked electronic - - PowerPoint PPT Presentation

UCL Institute of Health Informatics Big Data Science BAHIA 2018 7th–12th November 2018

The CALIBER Research Platform

Using large-scale linked electronic health records for research

Dr Arturo González-Izquierdo

University College London Institute of Health Informatics

• Data generation mechanism
• Linked electronic health records
• EHR phenotyping
• Challenges & opportunities

Healthcare system

General practitioners Hospitals Outpatients Specialists

Data generation mechanism

Initial point of consultation Health history Baseline characteristics Health behaviour Tests Medication Admitted patient care Acute events Diagnoses Procedures Referrals Elective patient care Monitoring Examination Emergency presentations Symptoms Signs Specialised consultative care Advanced medical investigation Specialised treatment & interventions

Healthcare settings and data custodians

CPRD: GP Data NHS Digital: Hospital Data Disease Registries: Tertiary care data ONS: Mortality Data

Electronic Health Records

Clinical Practice Research Datalink Office for National Statistics National Cancer Registration and Analysis Service Hospital Episode Statistics

Chronic cough Weight loss Chronic obstructive airway disease Recurrent mild chest infection Pneumonia Lung Cancer Diagnosis

Linked Electronic Health Records

PRIMARY CARE CANCER REGISTRATIONS SECONDARY CARE DEPRIVATION AND MORTALITY

Linked Electronic Health Records

Chronic cough Weight loss Chronic obstructive airway disease Recurrent mild chest infection Pneumonia hospitalisation Lung Cancer Diagnosis

Registration date Consultation date Consultation date Admission date Admission date Date of death Date of birth Blood tests (routine) Sputum tests Diagnosis Diagnosis (cancer type, stage, metastases) Underlying cause Blood pressure Spirometry Diagnosis Additional diagnosis (deep vein thrombosis Pneumonia) Procedures (surgery) Subsidiary causes Weight Chest x-rays Procedures Procedure for biopsy

• f lesion (bronchoscopy,

Chest drain) Treatment (chemotherapy, radiotherapy) Height CT chest Chest imaging (x-ray, CT, PET CT, chest drain insertion) Physical activity Treatment (antibiotic, Inhalers) Discharge date Health history (heart, diabetes, stroke) Smoking Alcohol Contraception Immunisations Discharge date

Linked Electronic Health Records

Chronic cough Weight loss Chronic obstructive airway disease Recurrent mild chest infection Pneumonia hospitalisation Lung Cancer Diagnosis

Registration date Consultation date Consultation date Admission date Admission date Date of death Date of birth Blood tests (routine) Sputum tests Diagnosis Diagnosis (cancer type, stage, metastases) Underlying cause Blood pressure Spirometry Diagnosis Additional diagnosis (deep vein thrombosis Pneumonia) Procedures (surgery) Subsidiary causes Weight Chest x-rays Procedures Procedure for biopsy

• f lesion (bronchoscopy,

Chest drain) Treatment (chemotherapy, radiotherapy) Height CT chest Chest imaging (x-ray, CT, PET CT, chest drain insertion) Physical activity Treatment (antibiotic, Inhalers) Discharge date Health history (heart, diabetes, stroke) Smoking Alcohol Contraception Immunisations Discharge date

Diagnoses or procedures Biometrics, test results, time dependent thresholds Medication EHR phenotype Health care utilisation patterns

EHR phenotype

EHR phenotype

• Extraction – Algorithm (generic)

Pujades-Rodriguez M. (2016) Heart, 102:383-398

The CALIBER Research Platform

Patient Population Cohort identification methods Deep phenotyping algorithms Longitudinal clinical trajectories Precise temporal allocation of Exposures and outcomes

Challenges

Day-to-day challenges:

• 1. Comply with the data custodians directives on data

protection

• 2. Understanding the data generation mechanisms

1.Clinical practice 2.Recording of information 3.Coding

• 3. Connecting jargons from multiple disciplines

4.Understand the associated information governance

Challenges

EHR’s observation window

Start End

Relevant clinical event Exposure to factor of interest Outcome of interest Death

Opportunities

• Recent willingness by data custodians to research health

data using machine learning based methodologies

• Wide range of exploratory or hypothesis generation/test

studies

– Patient classification (Machine Learning sub-phenotyping) – Detailed healthcare utilisation patterns (multi-state trajectory flows) – Integration of data models – Sophisticated epidemiological/statistical methods computationally feasible for causal inference – EHR based decision/early-detection tools (automation)

The Data Lab

Natalie Fitzpatrick Data Science Facilitator n.fitzpatrick@ucl.ac.uk CALIBER portal https://www.caliberresearch.org/portal Denaxas Lab http://denaxaslab.org/